just saw this banger of a post appear on my home feed (for some reason i thought i had the sub this was posted in muted already).
this man, a “software engineer” who admits he’s never used generative ai, is posting a secondhand story about someone vibe coding a text editor ui in one session with almost zero constraints…and then acting like this debunks frontier models.
he has never tried it himself, the friend one shotted a ui with basically “make a reddit-like editor using stylex”. the model didn’t follow the library perfectly, made some ugly code, one button broke. his conclusion is “llms are non deterministic slot machines, coding isn’t solved, boosters are lying”.
this is the state of the anti ai software engineer discourse in 2026. they’re still testing these things like it’s gpt 3.5 and then high fiving each other when pure vibe coding produces slop. nobody serious claims you can one shot maintainable software with zero harness, zero examples, and zero taste. the actual claim is that competent engineers who know how to drive the tools are moving significantly faster, especially on ui, boilerplate, refactors, and exploration.
this post doesn’t disprove the technology, just proves the OP isn’t serious.